| description abstract | Snowplow trucks serve a crucial role in winter maintenance activities by removing, loading, and disposing of snow. A performance-monitoring and analysis process can assist transportation agencies in effectively managing snowplow trucks and maintaining the normal functioning of roadways. Previous literature suggests that most snowplow truck performance analysis is done through life-cycle cost assessment at the macro level to determine the optimal life cycle for the entire truck fleet. However, this can lead to a waste of resources and may incur bias due to ignorance of performance variations resulting from endogenous and exogenous features. More important, such analysis fails to identify the factors that contribute to performance deterioration. With the proliferation of operational data on snowplow operations, these concerns can be addressed through predictive machine-learning (ML) techniques in a data-driven fashion. In this study, we applied a popular ML technique, random forest (RF), to predict the performance of snowplow trucks, which was quantified via the rank of major repair times. Another ML technique, linear support vector machine (SVM), was also applied for benchmarking and comparison. Using the snowplow truck fleet managed by the Utah Department of Transportation (UDOT), both models were implemented and it was demonstrated that RF outperforms linear SVM with regard to prediction accuracy. Further, a feature importance analysis from the RF model can help transportation agencies to improve truck replacement strategies by identifying crucial performance factors. Lastly, a sample application of the developed prediction model using RF suggests the threshold of work intensity for preventing the rapid deterioration of truck performance in various working environments. Compared with the life-cycle cost analyses used in previous studies, the prediction model proposed here can help transportation agencies to better prioritize fleet replacement. | |
